The human microbiome - the trillions of bacteria, viruses, and fungi that live in and on the human body - has emerged as a major area of cancer research. The microbiome influences cancer risk and outcomes through several mechanisms: promoting or suppressing inflammation, altering the genomic stability of host cells, producing metabolites that can either damage DNA or support immune function, and modifying how the body responds to cancer treatments including chemotherapy and immunotherapy.
Endometrial cancer is the most common gynecologic cancer in the United States, with increasing incidence, yet it lacks any routine blood or tissue-based screening test. It is classified into two broad categories: Type I (low-grade) tumors, which are estrogen-driven, typically endometrioid in histology, and often diagnosed early with favorable outcomes; and Type II (high-grade) tumors, which include serous, clear cell, carcinosarcoma, and high-grade endometrioid types, are more aggressive, often diagnosed at later stages, and carry worse survival.
Prior research has shown that the vaginal microbiome mirrors the microbiome of the upper genital tract (uterus and fallopian tubes) in women with cancer, suggesting that vaginal sampling could serve as a non-invasive proxy for studying uterine microbial environments. This study is one of the first to systematically examine whether the vaginal microbiome differs not just between benign disease and cancer, but between different cancer grades and histologic subtypes.
Vaginal swabs were collected from 61 racially and ethnically diverse patients at the time of hysterectomy surgery, before any antiseptic preparation of the surgical site (to avoid altering the native microbiome). Participants were divided into three groups: 11 controls with benign gynecologic conditions (such as fibroids), 30 low-grade endometrial carcinoma patients, and 20 high-grade endometrial carcinoma patients. The study was designed to oversample high-grade and serous carcinomas specifically, as these rarer, aggressive subtypes are underrepresented in most studies.
Microbial DNA was extracted from the swabs and subjected to shotgun metagenomics sequencing - a method that sequences all DNA in the sample, not just bacterial marker genes. Unlike 16S rRNA gene sequencing (the most common microbiome profiling method), shotgun metagenomics provides species-level (rather than genus-level) resolution and also enables analysis of microbial gene content and metabolic pathways - a significant methodological advantage for this type of study.
From approximately 7.1 billion total sequenced reads, 95.1% were human and removed from analysis. The remaining non-human sequences were taxonomically classified to identify 237 bacterial species. Microbial diversity was quantified using alpha diversity (how many distinct species are present per individual sample) and beta diversity (how similar or different the microbial communities are between samples). Machine learning models were then built to assess whether microbial abundance patterns could predict disease category.
Both alpha diversity (within-sample species richness) and beta diversity (between-sample compositional dissimilarity) were significantly associated with patient tumor grade. Specifically, high-grade endometrial cancer was associated with significantly greater microbial diversity compared to benign controls (p adjusted = 0.025), while low-grade cancer did not significantly differ from either benign controls or high-grade cancer in alpha diversity.
This pattern - higher diversity in more aggressive disease - is biologically interesting. A healthy vaginal microbiome is typically dominated by a few Lactobacillus species, which maintain an acidic environment that inhibits pathogen growth. Loss of Lactobacillus dominance and replacement by a more diverse community of bacteria (including anaerobes) is associated with conditions like bacterial vaginosis and elevated inflammatory states. The finding that high-grade cancer patients have more diverse vaginal microbiomes suggests that either the tumor microenvironment shapes the vaginal microbiome, or that microbiome dysbiosis contributes to creating conditions favorable to aggressive tumor development.
Crucially, beta diversity was significantly associated with tumor grade (p = 0.042) but NOT with race, ethnicity, age, or BMI - all potential confounders. This specificity to tumor-related factors, rather than demographic factors, strengthens the argument that the observed microbiome differences are genuinely related to the cancer biology rather than simply reflecting incidental differences in the patient populations.
Community state types (CSTs) are clusters of patients with similar vaginal microbial community compositions. Using hierarchical clustering, the study identified four distinct CSTs with significantly different microbiome compositions. Importantly, these CSTs mapped onto disease categories: benign patients predominantly clustered in CST1, low-grade cancer in CST2, and high-grade cancer in CST3 and CST4 (p = 0.036 overall). There was also significant variation in CST membership by histologic subtype (p = 0.017).
Differential abundance analysis identified specific bacterial species enriched or depleted in different cancer groups. Fusobacterium ulcerans and Prevotella bivia were significantly more abundant in high-grade compared to low-grade cancer. Conversely, Bifidobacterium longum - a species associated with immune protection and known to have low abundance in aggressive gastric cancer - was most depleted in high-grade disease. Fusobacterium nucleatum, a bacterium well-studied in colorectal cancer for promoting tumor growth and inducing chemotherapy resistance, showed more than 4-fold greater abundance in high-grade versus benign samples.
The high-grade cancer microbiome showed downregulation of microbial gene sets involved in homologous recombination, mismatch repair, and ABC transporters - pathways related to DNA repair and drug efflux. This is particularly intriguing because downregulation of DNA repair pathways in the tumor-associated microbiome could influence the tumor microenvironment and potentially affect treatment sensitivity, though the causal directionality of this association requires further investigation.
Random forest classifiers trained on microbial species abundance data demonstrated meaningful diagnostic performance. The model distinguishing tumor (all cancer) from benign achieved a mean prediction AUC of 0.878, using just 3 key species to make this distinction. This suggests that the microbiome difference between cancer and no cancer can be captured with a compact, potentially clinically deployable signature.
Two additional models addressed the more clinically challenging task of distinguishing cancer grades: the model distinguishing high-grade from benign achieved AUC 0.80, and the model distinguishing high-grade from low-grade cancer achieved AUC 0.77. The ability to distinguish low-grade from high-grade endometrial cancer preoperatively from a vaginal swab would have significant clinical value, as high-grade tumors require more aggressive surgical staging and adjuvant treatment.
Models based on tumor histologic subtype (e.g., serous vs endometrioid vs benign) showed AUC values of 0.776-0.826, with the benign-versus-serous carcinoma model performing best. These results confirm that the vaginal microbiome carries not just information about cancer presence, but about the specific biological subtype of cancer - information that currently requires surgical tissue sampling to obtain.
This exploratory study establishes that the vaginal microbiome contains clinically informative signals about endometrial cancer grade and histology - not just whether cancer is present, but what kind it is. A simple vaginal swab collected at the time of gynecologic evaluation could potentially provide microbiome-based information to complement standard clinical workup, representing a non-invasive and inexpensive window into uterine tumor biology.
The clinical implications are particularly relevant for distinguishing low-grade from high-grade endometrial cancer. Preoperative identification of high-grade tumors currently relies on endometrial biopsy - which can miss the most aggressive areas of the uterine lining due to sampling limitations - and MRI imaging. A microbiome signature that independently predicts high-grade disease could improve preoperative risk stratification and surgical planning.
Important limitations include the small sample size (61 patients), single-institution enrollment, and the cross-sectional design that prevents causal conclusions about whether microbiome changes precede or follow tumor development. Whether the observed associations would hold in larger, more diverse populations - including non-English speaking communities and women from different geographic regions with different baseline microbiome compositions - requires prospective multicenter validation. Future research should also explore whether modifying the vaginal microbiome through probiotics or targeted antimicrobial therapies could influence cancer outcomes.